DevOps-based Collaboration for Building and Operating SDN-enabled Multisite Cloud Playground

Abstract

Emerging paradigms, such as software-defined networking (SDN), cloud computing and network function virtualization (NFV), are the key motivators for building and operating a testbed infrastructure for Future Information and Communication Technology (ICT) services. Several testbed deployment efforts are being carried out worldwide with the enormous number of experiment users. Unfortunately, some of the deployments are very challenging due to limited of supporting resources and the heterogeneity of underlay network infrastructure. In this regard, we propose a concept of a playground (i.e., miniaturized testbed) which is software-defined to be provisioned only using software-based functions and continuously operated by tiny-sized DevOps-style team. It has centralized centers inside the playground tower, to control and monitor distributed hyper-converged box-style site resources. The goal is to have a lightweight but distributed infrastructure for service-oriented experiment with these three characteristics which are “affordable”, “visible”, and “federated”. In this specific work, we mainly address the affordability aspect of the playground by adopting hyper-converged Whitebox-style resources (i.e., SmartX Box) with a set of open-source-based software implementation to support automated provisioning and diverse experimentation (e.g., SDN, Cloud, and NFV). Furthermore, the playground should allow multiple developers for executing their experiments simultaneously across the geographically distributed resources. In this situation, flow-centric visibility becomes very important to address a limited visible characteristic of the playground by leveraging embedded tracing features from Linux-based box resources for clustering and identifying different types of flow. The tenant-based or behavior-based flow clustering and identifications are required to recommend specific action for a specific type of flow with the help of the centralized intent-leveraged networking control through the ​SDN controller. This networking control can apply the recommended action into the network or device policy, which is implemented by installing flow rules in the SDN-based switch. It is also extended to initially address the federated characteristic of the playground by providing a limited communication between multisite SDN controllers under multi-domain network administrators through Software-defined-routing-exchange (SDRE) mechanism. It modifies intent configuration based on the BGP IP routing information which is exchanged between the routers in the multisite playground. Based on the verification and measurement results, our proposed hyper-converged box-style resources is required less provisioning (i.e., installation or configuration) time, especially for the site with a good Internet connection, and also lower deployment cost compares to rack-style resources. Thus, tracing-based flow collection in the Linux-based box resources gives a lower CPU overhead process by applying packet tracing in the kernel-space. Finally, intent-leveraged networking control has been verified to establish underlay-aware multisite interconnection based on IP prefix, and also to redirect “unknown” flow from specific prefix. It also shows less performance differentiation for a large number of routes and implementation stability over the real BGP-based IP network.